Health Data Platform

Principal Investigator: Dr Will Innes

This proof of principle project will apply advanced Machine Learning computer algorithms to obtain deeper insights from the data already collected and to realise the value of the Trust’s vast healthcare data repository.

The Study

Brain diseases such as dementia affect many older people. These can be challenging for doctors to diagnose, especially in the early stages when treatment would be most effective. One reason is that doctors can’t examine the brain directly. Brain scanners help, but are too expensive for routine screening, and only produce coarse images which may not reveal early-stage disease. But there is one part of the central nervous system which can be examined directly – the retina at the back of the eye. New technology can now scan the retina quickly and cheaply, and in microscopic detail.

Our group has been investigating whether eye scans can reveal brain diseases like Parkinson’s and Alzheimer’s. If so, this could lead to better screening techniques which enable these diseases to be detected early.

Eye scans contain vast amounts of information, but so far we have used relatively simple ways of analysing them. We’re concerned that we may be missing important features. Fortunately, new computer techniques are incredibly good at spotting patterns in large amounts of data. We now want to apply these techniques to our eye scans, training computers to spot subtle signs of disease which are hard for humans to detect.

Of course, it’s essential that patients consent for us to use their data in this way, and that data be kept completely secure while it’s being analysed. One important aim of this project is to set up ways for the NHS, Newcastle University and patients to work together to ensure this happens.

Next steps

This project aims to overcome technical and governance hurdles and provide “proof of principle” regarding the application of Machine Learning techniques to large NHS data-sets. We expect application of these techniques to this very high quality dataset to provide pilot results to support subsequent larger grant applications (for example EPSRC Healthcare Technologies theme or to the future MRC-funded Health Data Research UK Institute) to develop advanced, non-invasive biomarkers for neurodegenerative diseases.

If this project is successful, we and colleagues across NUTH/NU will be able to access large NUTH clinical data-sets, containing tens of thousands of patients, alive and deceased, for Machine Learning research. These clinical records contain data such as haematology, electrocardiology, neuro and bone radiology and (once the GCDE is complete) prescriptions, added to additional data collected prospectively as part of other research programmes (such as electrodiagnostics, ophthalmic imaging or Positron Emission Tomography data). These rich data-sets offer the potential for earlier diagnosis and better treatment as well as mechanistic insights across a very wide range of clinical conditions.

Our long-term goal is to help position Newcastle to be a leader in medical informatics and to enable NUTH to extract value from its store of raw digital healthcare information.

Project milestones

• Establish clear governance arrangements and standards within trust and university.
• Complete technical development for data transfer, storage and manipulation within university.
• Establish accuracy of automated diagnosis from classifier based on supervised learning.
• Establish features extracted from data by unsupervised learning algorithm.